Claude CodeEngineer
Find our deployment doc in the wiki and take this project through deploy
wiki_search("deployment steps")
Deployed — every step linked to its source.
OTOntology
Still pasting the same doc?
Every new chat window starts from zero on what your company knows.
Switch tools, paste again. Open a window, paste again. Every day, in every tool.
Talk to us
Connecting is the whole job
Ask, and it answers with sources.
WHAT YOU GET
This is the knowledge layer
A company wiki you can ask, with a link to the original on every page.
JUST CONNECT
Build it yourself, then maintain it forever
Connect once. We handle the rest
Nothing to build. Connect your tools, and scattered documents become a wiki your team can ask.
- Checks for changes every 5 minutes
- Rebuilds only what changed
ASK IN SLACK · DISCORD
Answer the same question, again
Ask where you already work. Just ask
Mention the bot and it searches the knowledge layer for evidence. Nobody answers the same question twice.
- Slack
- Discord
- No new tool to learn
NO EVIDENCE, NO ANSWER
Sounds right, cites nothing
No evidence, no answer
When the search returns nothing, we never call the model. It gets no chance to invent.
- Every answer ships with its source
How the refusal works
MCP · ANY AGENT
Any AI tool. One address
Finding is table stakes. Your agents finish the work.
Once connected, this is how you use it
claude mcp add --transport http otontology https://otontology.otoworks.ai/mcp- Claude Code
- Cursor
- Codex
- Gemini CLI
- Windsurf
- Anywhere MCP runs
What agents actually do with it
CursorPM
Draft the kickoff doc for the search revamp, including why we chose this structure
wiki_search("OKR review")
Draft ready — decisions quoted from the meeting notes.
ClaudeSupport
Draft a reply to this customer using our support manual
wiki_fetch("support manual")
Draft ready — manual linked inline.
AGENTIC RETRIEVAL
Pile it on the server, or just click and connect?
AS-IS
Server-side — the whole pipeline lives on the server
Steps the server owns: 6+
- Question
- Query planning
- Sub-query split
- Iterative search
- Reranking
- Self-evaluation
- Synthesis
- Answer
And all of it is server ops —
"reasoning effort" tuningIndex pipeline upkeepPermission syncEval dashboardsCost & latency monitoringTO-BE
Client-side — the server only has to search well
What the server does: search / fetch. That's it
QuestionAnswer
Client agent
Claude Code · Cursor
Planning, splitting, iterating happen here
search / fetch
Claude Code · terminal
claude mcp add --transport http otontology https://…/mcp
Connected — the agent works with your team's knowledge in hand
One address line. Click, connected — leave the hard parts to the agent that already does them well; the server focuses on search quality.
AUDIT LOG
Every answer leaves a trail
Who asked, what came back, and which documents it drew from. Human questions and agent lookups alike.
Isolated per company
Each company's data stays separate. Another company's documents never enter your search.
Agents read only
What we open to agents is read-only. They can't change the wiki, the index, or the catalog.
Lookups are logged
Human questions and agent lookups land in the same log. You can check what any answer was based on.
WHY KNOWLEDGE LAYER
The base layer of the AI-agent era: the knowledge layer
For an agent to work, it has to read what the company knows. The industry calls this the knowledge layer. It gathers scattered company knowledge in one place so people and AI read the same evidence. OTOntology builds that layer from a Notion, Slack, or Discord connection alone.
“Why Enterprise AI Starts With A Knowledge Layer”
Connect, and the knowledge layer builds itself.
One answer for people and for every AI.
No evidence, no answer.
A knowledge layer that answers with sources
Reach out and we'll set up your workspace connections with you.
Talk to us